Unity ML - Learning during gameplay

Hi.

I'm pretty new to unity but I work with ML for decades. I loved the ML-Agents tools and it gave me several ideas.

For now, I'm learning Unity and refreshing my C# skills

One thing I'm trying to find is if there is a way to train the ML model during gameplay in a published project (preferably mobile) and use that learning to update the agents.

For example, let's say I have a standard tower defence game. As the player sets up the towers, I would like to have enemies that "learn" the pattern and in the future waves are able to avoid the turrets automatically (I know it is not hard to achieve this through heuristics, just one example). From every tutorial and documentation I have found, I would have to pre-train the models during development and publish the trained NN with the game.

Is there a way to train and update the NN during gameplay? Please, do you have a link for a tutorial/example?